Research map / Label-efficient and robust learning / Domain adaptation and generalization
Associative Domain Adaptation
Philip Haeusser, Thomas Frerix, Alexander Mordvintsev, Daniel Cremers
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add association loss to source-only training | generic CNN · MNIST -> MNIST-M | target test error (%): 35.96 -> 10.53 | Improved |
| add association loss to source-only training | generic CNN · Synth Digits -> SVHN | target test error (%): 15.68 -> 8.14 | Improved |
| add association loss to source-only training | generic CNN · SVHN -> MNIST | target test error (%): 30.71 -> 2.40 | Improved |
| add association loss to source-only training | generic CNN · Synth Signs -> GTSRB | target test error (%): 4.59 -> 2.34 | Improved |
| use association loss instead of MMD loss as similarity term (ablation) | generic CNN · MNIST -> MNIST-M | target test error (%): 22.90 -> 10.53 | Improved |
| use association loss instead of MMD loss as similarity term (ablation) | generic CNN · Synth Digits -> SVHN | target test error (%): 19.14 -> 8.14 | Improved |
| use association loss instead of MMD loss as similarity term (ablation) | generic CNN · SVHN -> MNIST | target test error (%): 28.48 -> 2.40 | Improved |
| use association loss instead of MMD loss as similarity term (ablation) | generic CNN · Synth Signs -> GTSRB | target test error (%): 10.69 -> 2.34 | Improved |
| add association loss on top of source-only training vs MMD loss on Synth Digits -> SVHN (ablation) | generic CNN · Synth Digits -> SVHN | target test error (%): 15.68 -> 19.14 | Hurt |
| add association loss with fixed hyperparameters on Synth Signs -> GTSRB | generic CNN · Synth Signs -> GTSRB | target test error (%): 4.59 -> 17.20 | Hurt |
| remove batch size constraint on traffic signs (ablation) | DA_assoc fixed params · Synth Signs -> GTSRB | target test error (%): 17.20 -> 6.55 | Improved |
| add association loss instead of MMD loss, in terms of embedding MMD | generic CNN · SVHN -> MNIST | MMD between embeddings: 0.0404 -> 0.2112 | Hurt |
Closest papers by meaning
- Active Universal Domain Adaptation (ICCV 2021 · similarity 0.90)
- Domain-Adversarial Training of Neural Networks (JMLR 2016 · similarity 0.90)
- Minimum Class Confusion for Versatile Domain Adaptation (ECCV 2020 · similarity 0.90)
- Boosting Domain Adaptation by Discovering Latent Domains (CVPR 2018 · similarity 0.90)
- Category-Aware Active Domain Adaptation (ICML 2024 · similarity 0.89)
- Universal Semi-Supervised Domain Adaptation by Mitigating Common-Class Bias (CVPR 2024 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
